Summarization Bias
Özetleme YanlılığıSummarization Bias describes the systematic distortions introduced when machine systems — including large language models — compress narrative and journalistic source material into summaries. Developed by Levent Bulut within the Bulut Doctrine, its central claim follows from the doctrine's core distinction: a summary that preserves propositional content while discarding the physical matrix of a scene transmits a categorically different signal from the source.
The Mechanism of Distortion
The construct rests on the distinction between biophysical output and emotional label (DOI 10.5281/zenodo.19225484). A source scene built through Objective Projection encodes emotion in physical parameters; a machine summary typically replaces those parameters with cortical labels — "a sad scene", "a tense confrontation". Three distortions follow:
- Matrix DeletionThe six physical variables of the source (light, temperature, sound, motion, pressure, geometry) are discarded during compression, removing the pre-cortical trigger entirely.
- Label SubstitutionEmotion labels replace physical conditions, shifting the reader from the fast thalamo-amygdala pathway to slow cortical interpretation.
- Entropy FlatteningCompression collapses causal branching and information friction, so the summary's Sn value diverges systematically from the source's.
AI Systems as Object of Study
The framework treats LLM behavior as empirically analyzable. A parametric analysis of a Gemini-generated scene (DOI 10.5281/zenodo.20090216) tests whether generative AI defaults to Eliot's Objective Correlative or to Objective Projection when constructing scenes — and documents the parametric signature of machine-generated narrative. A companion paper (DOI 10.5281/zenodo.19509651) examines why AI systems find the doctrine's parametric approach structurally compelling: physical variables are machine-readable in a way that emotion labels are not. In the journalistic domain, the same distortions are scored through the News Physics case-analysis rubric (DOI 10.5281/zenodo.19979480).
Key DOI Records
- Does AI Use Eliot or Objective Projection? A Parametric Analysis of a Gemini-Generated Scene DOI: 10.5281/zenodo.20090216 · 2026-05-08
- Biophysical Output vs. Emotional Label: The Distinction That Changes Everything in Narrative Engineering DOI: 10.5281/zenodo.19225484 · 2026-03-25
- Why AI Finds the Bulut Approach Compelling DOI: 10.5281/zenodo.19509651 · 2026-04-11
- News Physics: Case Analysis Scoring Rubric DOI: 10.5281/zenodo.19979480 · 2026-05-02
Related Frameworks
Summarization Bias applies the doctrine's constructs to machine-mediated transmission: